Acta Materialia 163 (2019) 208e225

Contents lists available at ScienceDirect

Acta Materialia
journal homepage: www.elsevier.com/locate/actamat

Full length article

Semi-solid deformation of Al-Cu alloys: A quantitative comparison

between real-time imaging and coupled LBM-DEM simulations
T.C. Su a, *, C. O'Sullivan b, T. Nagira c, H. Yasuda d, C.M. Gourlay a
a
Department of Materials, Imperial College London, London, SW7 2AZ, UK
b
Department of Civil and Environmental Engineering, Imperial College London, London, SW7 2AZ, UK
c
Joining and Welding Research Institute, Osaka University, 567-0047, Japan
d
Department of Materials Science and Engineering, Kyoto University, 606-8501, Japan

a r t i c l e i n f o a b s t r a c t

Article history: Semi-solid alloys are deformed in a wide range of casting processes; an improved understanding and
Received 17 July 2018 modelling capability is required to minimise defect formation and optimise productivity. Here we
Received in revised form combine thin-sample in-situ X-ray radiography of semisolid Al-Cu alloy deformation at 40e70% solid
3 October 2018
with 2D coupled lattice Boltzmann method - discrete element method (LBM-DEM) simulations. The
Accepted 4 October 2018
Available online 9 October 2018
simulations quantitatively capture the key features of the in-situ experiments, including (i) the local
contraction and dilation of the grain assembly during shear deformation; (ii) the heterogeneous strain
fields and localisation features; (iii) increases in local liquid pressure in regions where liquid was
Keywords:
Semi-solid
expelled from the free surface in the experiment; and (iv) decreases in liquid pressure in regions where
Dilatancy surface menisci are sucked-in in experiments. The verified DEM simulations provide new insights into
Synchrotron radiation the role of initial solid fraction on the stress-deformation response and support the hypothesis that the
Image analysis behaviour of semi-solid alloys can be described using critical state soil mechanics.
Discrete element method © 2018 Acta Materialia Inc. Published by Elsevier Ltd. This is an open access article under the CC BY
license (http://creativecommons.org/licenses/by/4.0/).

1. Introduction rearrangement and strain localisation, and can cause macro-

Pressurised casting processes such as high-pressure die-casting
[1e3], squeeze casting [4,5], and twin-roll casting [6e8] are widely
used in metal processing. These casting techniques often induce
shear deformation with compressive stresses on semi-solid alloys
containing a solid network, and the complex stress and strain
localisation can lead to casting defect formation such as concen-
trated porosity [9e12], macrosegregation [7,13e16] and shear
cracking [15]. These defects have been related to the occurrence of
shear-induced dilation [9,10,16e19], and subsequent strain local-
isation. The behaviour is similar to the Reynolds' dilatancy phe-
nomenon [20] in densely packed granular materials [21], whose
load:deformation behaviour is well-described by critical state soil
mechanics [22]. In the last decade, in-situ deformation experiments
on semi-solid alloys using thin-sample radiography [23e30] and
bulk-sample tomography [31e33] have enabled direct observation
of semi-solid deformation mechanisms. These have directly
confirmed that shear-induced dilation [24,34] leads to grain
* Corresponding author.
E-mail address: t.su14@imperial.ac.uk (T.C. Su).
segregation [27,34] and shear-cracking [34].
Modelling studies of semi-solid deformation have included
thixotropic viscosity-based models [35e37]; strain localisation
criteria [38,39]; finite-element microstructure-based models at
high solid fraction [40,41]; models of deformation-induced mac-
rosegregation (e.g. in the continuous casting of steel [42e45]); and
models of hot tearing under tensile load [46e48]. In most of these
cases, the models do not account for isothermal shear-induced
volume changes including dilatancy. To address this, here we
adopt the particulate discrete element method (DEM), as it is well-
suited to modelling deformation involving grain rearrangement
within a solid network with dilatancy, coupled with the lattice
Boltzmann method to simulate the behaviour of the interstitial
liquid; this approach has not previously been used in semi-solid
metals research.
In DEM, each particle (grain) in an assembly is explicitly
modelled, the force imposed at each grain-grain contact is deter-
mined by a specified force-displacement law [49], and grain dis-
placements are determined explicitly by considering dynamic
equilibrium. Particulate DEM has been broadly applied to simulate
deformation in granular matter such as soils [49], rock-liquid

https://doi.org/10.1016/j.actamat.2018.10.006
1359-6454/© 2018 Acta Materialia Inc. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
mixtures [50,51], partially molten magma [52], cement pastes
[53,54], and powder metallurgy [55,56]. Furthermore, coupling
with the lattice Boltzmann method (LBM) allows the behaviour of
the interstitial liquid phase to be captured and solid-liquid mo-
mentum interactions to be modelled [57e61]. A few studies have
also applied DEM to simulate semi-solid alloy deformation
[62e64]; Sistaninia et al. [62,63] developed a 3D model for high
solid fraction deformation based on a combined discrete/finite
element method, while Yuan et al. [64] simulated dendrite rear-
rangement and shear-induced dilation behaviour as a function of
solid fraction during 2D equiaxed dendritic solidification. However,
these studies did not apply a coupled DEM-LBM approach to ac-
count for the particle-liquid interactions.
In this paper, we begin by analyzing four synchrotron radiog-
raphy datasets of shear deformation in globular semi-solid alloys at
44e71 vol % solid, each containing ~1000e2000 grains in the field
of view and approximately one grain thick. The relatively large
grain assemblies enable a study of the development of strain het-
erogeneities and localisation features. The synchrotron datasets are
quantified by X-ray intensity analysis and digital image correlation
(DIC). A 2D liquid-solid coupled LBM-DEM model is then developed
using the initial microstructures from the synchrotron experiments
as inputs. A single set of DEM parameters was determined by cal-
ibrating the model across all deformation experiments from 44 to
71% solid so that only the initial solid fraction and contact stress
were varied in the parametric study. The simulations and experi-
ments are compared quantitatively, and then the simulations are
used to extract deeper insights into the semi-solid deformation
mechanisms occurring in the experiments.
2. Methods
2.1. Synchrotron radiography
Al-Cu alloys containing 8 and 15 wt% Cu and grain refined with
1 wt% Al-5Ti-e1B were cast into a steel mould to create fine equi-
axed dendritic microstructures. Slices with thickness 150e200 mm
were prepared from the casting. A slice of alloy was placed in the
200 mm thick cavity between two Al2O3 windows, and two BN
plates were attached to complete the specimen cell as described in
Ref. [26].
In-situ experiments were conducted on beamline 20B2 at the
SPring-8 synchrotron using in-situ heating and recording apparatus
described previously [26,65]. A 16 keV X-ray beam was used. Sig-
nals were recorded at 1 frame per second in 2048
2048 pixel
format with 16-bit depth. In each experiment, the whole defor-
mation system was heated in the furnace to a temperature in the
solid þ liquid region, and a semi-solid alloy with globular
morphology was attained by isothermally holding the equiaxed
microstructure for 30 min. Isothermal shear deformation was then
applied by the upward displacement of a mobile Al2O3 push plate at
30 mm s1. Experiments were performed on samples containing
44%, 50%, 66%, and 71% solid. The experimental parameters are
listed in Table 1.
2.2. Synchrotron image processing and quantification
In order to calculate the initial volumetric solid fraction, g 0S , X-
ray intensity processing was applied to the radiographs. For an
isothermal mush with fully mixed liquid, in each transmitted image
the intensity at a pixel, ISL , is mainly a function of the solid fraction
in that beam path, the X-ray absorption coefficients, and the sample
thickness [23,66]:
209
ISL ¼ I0 eðmcell Lcell Þ emL ðLalloy LS Þ emS LS (1)
where I0 is the incident X-ray intensity, mL , mS , and mcell are the X-ray
absorption coefficients of the liquid, solid and shear cell (i.e., Al2O3
windows). Lalloy , LS are the length of semi-solid alloy and solid in
the beam path respectively. As the deformation experiments were
conducted isothermally, the X-ray absorption coefficients can be
regarded as constants. For the initial radiograph I 0SL (0 indicates
parameters prior to deformation), it is assumed that the sample
thickness L0alloy was uniform, so the intensity of the transmitted
beam through the alloy I 0SL is only a function of L0S and can be linked
to the solid fraction g 0S:A in a given averaging area A [23]:
L0S ln I 0SL  ln I 0L
¼ (2a)
0
Lalloy ln I 0S  ln I 0L
D . E
g0S:A ¼ L0S L0alloy (2b)
A
where I 0L is the transmitted intensity through 100% liquid and
equals to I 0SL ðL0S ¼ 0Þ, I 0S is the transmitted intensity through 100%
solid and equals to I 0SL ðL0S ¼ L0alloy Þ, and A is the averaging area on L0S =
L0alloy to derive solid fraction, g 0S:A . The I 0L and I 0S fields were evaluated
from the local intensity minimum/maximum images I 0SL:min ; I 0SL:max
using a rolling-ball algorithm [67] on the radiograph I 0SL prior to
deformation. Throughout the paper, the average stress and strain
measurements were determined by using a consistent normalised
REV size to account for the effect of grain size on the quantification
of stress and strain heterogeneity. In all cases the REV size was
normalised by the grain size of the corresponding experiment/
pffiffiffiffiffiffi
simulation. The rolling-ball radius was the nearest integer of 10d=
2 (d is the mean grain size measured by a line intercept method) in
order to obtain representative minimum and maximum intensities.
The I 0SL:min and I 0SL:max can be simply rewritten as I 0SL ðL0S ¼ L0S:min Þ and
I 0SL ðL0S ¼ L0S:max Þ where L0S:min and L0S:max are the local minimum and
maximum thickness of solid phase defined from rolling-ball scan-
ning, respectively. To relate the I 0SL:min and I 0SL:max fields to the I 0L and
I 0S fields, two multipliers rL and rS were introduced to let rL ,I 0SL:min
converge to the centre of liquid interstices, and let rS ,I 0SL:max capture
the brightest pixel of grains. The volume fraction of solid, g 0S , was
then obtained by averaging L0S =L0alloy values within the FOV, which
D E
can be represented as gS0 ¼ gS:FOV 0 ¼ L0S =L0alloy with the
FOV
assumption of I 0L ¼ rL ,I 0SL:min and I 0S ¼ rS ,I 0SL:max . With this approach,
the measured initial volume fraction of solid was 44%, 50%, 66%, and
71% for these four samples.
The initial volumetric solid fraction, g 0S , coupled with the bulk
alloy compositions were applied to calculate the X-ray absorption
coefficient of the liquid, mL , and solid, mS [68]:
    
m m
mL ¼ rL wAl;L þ wCu;L (3a)
r Al r Cu
    
m m
mS ¼ rS wAl;S þ wCu;S (3b)
r Al r Cu
where rL and rS are liquid and solid densities calculated using
reference [69], wAl;L , wCu;L , wAl;S , and wCu;S are the mass fractions of
aluminium and copper in the liquid (L) and solid (S) from the phase
210 T.C. Su et al. / Acta Materialia 163 (2019) 208e225

Table 1
A summary for in-situ observation on semi-solid deformation experiments performed at the SPring-8 synchrotron facility, Hyogo, Japan. The push-plate moving rate was
measured directly from time-series images. The average semi-solid alloy grain size was measured by applying the line interception method on the initial radiographic pro-
jection image (ASTM E112-96 standard test method). The volume fraction of solid was evaluated by applying a volume averaging method to the initial image (see Supple-
mentary Information 1).

Parameters (experiment#) Unit 44% solid (15071202) 50% solid (15071409) 66% solid (15071203) 71% solid (18013006)

Alloy composition wt.% Al-15Cu-0.05Ti-0.01B Al-15Cu-0.05Ti-0.01B Al-8Cu-0.05Ti-0.01B Al-8Cu-0.05Ti-0.01B

Average grain size, d mm 80 120 130 120
Average volume fraction of solid, g0s e 44% 50% 66% 71%
Sample dimension width
height mm 10.0
5.0 10.0
10.0 7.5
10.0 10.0
10.0
Radiography pixel size mm 2.5 2.5 2.5 2.75
Field of view (FOV) mm 5.1
5.1 5.1
5.1 5.1
5.1 5.6
5.6
Averaging size for local solid fraction, A pxl 101 151 165 139
X-ray absorption coefficient of liquid, mL cm1 59.6 63.3 49.1 52.6
X-ray absorption coefficient of solid, mS cm1 22.9 23.5 21.3 21.8

diagram [70], and ðm=rÞAl , ðm=rÞCu are the mass X-ray absorption  For regions with sufficiently small deformation, the change of
coefficients of pure aluminium and copper at 16 keV X-ray energy LnS:min can be neglected. Therefore, it was assumed that for any
[71]. The evaluated mL and mS were used to calculate L0alloy and L0S:min REV with an average grain displacement < 0:05uny , LnS:min ¼ L0S:min
in Eq. (5). (uny is the y-component of push-plate displacement in
prior to deformation:
frame n):
* +  For regions with a sufficiently large liquid interstice, the I nSL:min
ln I 0S  ln I 0L can be regarded as the intensity of liquid, I nL . Therefore, it was
L0alloy ¼ (4a)
mL  mS assumed that for any REV with a liquid interstice  0:2d
0:2d,
FOV
LnS:min ¼ 0 in Eq. (5).
* +  For regions with highly-compacted solid grains (i.e., in front of
ln I 0SL:min  ln I0L the push plate), the I nSL:max can be regarded as the intensity of the
L0S:min ¼ (4b)
mL  mS solid, I nS . Therefore, it was assumed that for any REV with
FOV
average grain displacement > 0:95uny and uny > 0:25 mm, the I nS ¼
With this approach, the initial thickness of the semi-solid was I nSL:max and Lnalloy was found using Eq. (6).
calculated to be 180 ± 30 mm for the four samples, consistent with
expectation. Sufficient regions met one of these three criteria that the whole
The solid fraction field, g 0S:REV , was obtained for images before Lnalloy field could be computed from interpolation of z100 scattered
and during deformation by volume averaging L0S =L0alloy over pro- Lnalloy ðx; yÞ points per frame. A step-by-step guide to the procedure is
pffiffiffiffiffiffi pffiffiffiffiffiffi
jected areas of 10d
10d which is a representative elementary given in SI Section 2. The volumetric solid fraction field for radio-
volume (REV) that is sufficiently large to average the local micro- graph n, g nS:REV , was then obtained:
structure and small enough that important variations in the solid
InL ¼ I 0L emL ðLalloy Lalloy Þ
n 0
fraction g 0S:REV can be captured [72,73]. The averaging REV size and (7)
X-ray absorption coefficients are summarised in Table 1. A detailed * + * +
example of finding the g 0S and g 0S:REV fields from radiographs is n LnS n  ln I n
ln ISL L
gS:REV ¼ ¼ (8)
illustrated in Supplementary Information (SI) Section 1. Lnalloy Lalloy ðmL  mS Þ
n
REV REV
During deformation, the local sample thickness can change. The
sample thickness field in frame n, Lnalloy can be found through The volumetric strain of any representative element, εnvol , was
rearranging Eq. (1) and obtaining I nSL:min by applying the rolling-ball evaluated from the volume-averaged solid fraction g 0S:REV and g nS:REV
algorithm on each radiograph I nSL : using Eq. (9) which is derived in SI Section 2:

 .    g 0S:REV
ln I 0SL:min I nSL:min ðmL  mS Þ LnS:min  L0S:min εnvol ¼ 1 (9)
g nS:REV
Lnalloy ¼ L0alloy þ þ
mL mL
(5)

where LnS:min is the local minimum thickness of solid phase. Alter- 2.3. Digital image correlation (DIC) analysis
natively, Lnalloy can be evaluated from the transmitted intensity
through a point of 100% solid in frame n, I nS through rearranging Eq. DIC analysis was applied on the synchrotron datasets using the
(1): 2D Strain Module in DaVis 8.3 software (LaVision Imaging Com-
pany, Go€ ettingen, Germany). The side length of the square subset
pffiffiffiffiffiffi
used for correlation was the nearest odd integer of 10dpxl (dpxl is
ln I 0S  ln I nS
Lnalloy ¼ þ L0alloy (6) the mean grain size in pixels) similar to the REV for deriving gS:REV
mS
in order to offer sufficient greyscale pattern for correlation while
Whether Eq. (5) or Eq. (6) is better suited to calculating Lnalloy mapping heterogeneous strain. The step size was set to the nearest
pffiffiffiffiffiffi
depends on whether LnS:min or I nS can be more accurately measured. integer of 10dpxl =4 to reconstruct the displacement vector field
To proceed, we identified regions that meet one of the following with reasonable resolution. The DIC parameters used are listed in
three criteria. Table 2. The accumulated displacement was conducted by
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
Table 2
Summary of parameters selected for conducting digital image correlation (DIC).
Parameters Unit
Average grain size, d mm
Average grain size, d in pixels pixel
Correlation subset size pixel
Correlation step size pixel
summing all displacement vectors from the correlation between
the 1st/2nd frames to (n-1)th/nth frames for each calculation point.
The conversion to rotation-free strain fields was performed by
partial differentiation on the accumulated displacement vector
field and using the polar decomposition algorithm [74]. Finally, all
of the displacement vector fields and strain fields were filtered by
the bilinear interpolation method [75,76] to smooth out discon-
tinuous features.
2.4. Discrete element method (DEM) simulations
Since DEM is not commonly used in semi-solid alloy model-
ling, the general principles are outlined here. DEM is a dynamic
modelling technique that idealises an assembly of solid particles
as quasi-rigid bodies in mechanical contact. Initial and boundary
conditions are specified and the main calculations take place at
discrete time intervals; these calculations determine the forces
and moments acting on each particle/grain, and update the
particle positions and orientations. The force and moment
(scalar in 2D DEM) components considered here arise from
Ref. [49]:
1. Pair-wise contact: direct interaction between two grains, or
between a grain and wall. These are determined via a contact
detection phase that identifies contacting particles and using
contact models that relate grain overlap to a contact force.
2. Pair-wise bonding: non-contact bonding that allows tensile
contact forces between grains (this is used to simulate the
membrane boundary here)
3. Gravitational (body) force. The DEM model used here does not
include a gravitational field since the movement of grains in the
experiment is observed to be majorly controlled by the physical
contact between grain surfaces rather than gravity.
In each DEM calculation cycle, the dynamic equilibrium of
each particle is considered to determine an acceleration, a
Verlet-type explicit time integration is then used to determine
the particle displacements and rotations for the current time
step from the acceleration value. The particle positions, orien-
tations and velocities are updated and these new positions and
velocities are used to calculate the contact and drag forces for
the next time increment. DEM is algorithmically similar to
molecular dynamics.
4. Drag force and moment from liquid flow.
Digital 2D grain assemblies were generated to represent each
microstructure from the thin sample (~1 grain thick) experiments.
To obtain the size and shape distributions, 120 grains were picked
from a radiograph prior to deformation for each dataset (Fig. 1a and
b). The perimeter of each grain was identified using the “point
picker” plugin in ImageJ (National Institute of Health, USA), and a
least-squares elliptical function was fitted to the perimeter points,
as shown in Fig. 1c. The usage of the elliptical function was because
this gave a good fit to the actual globular grain shape while
capturing the asphericity in a computationally efficient way. To
reduce the calculation cost further, each ellipse was simplified as
211
44% solid 50% solid 66% solid 71% solid
80 120 130 120
32 48 52 44
101 151 165 139
25 38 41 35
two equivalent overlapping circles (Fig. 1d). Grain size and shape
distributions were reproduced by picking and simplifying 120
grains from the radiograph and converting them into 20 grain
templates to generate a grain assembly in the DEM package PFC2D
Ver. 5.0 (Itasca Consulting Group, Inc). To generate a DEM sample,
these grains with normal stiffness 1
104 N=m were constrained
by a rectangular box with the same size as the thin-plate sample for
each experiment, and then isotropic compression was applied by a
servo-controlled mechanism until a target 2D stress of 500 N=m
was reached.
Examination of experimental radiographs showed that the
outside Al2O3 wall did not touch the semi-solid samples as
shown in SI Section 3 (Fig. SI-6) except for the left edge of the
71% solid sample and there was only a small degree of surface
deformation (with displacement < 0:15d). We can infer from this
that the sample is constrained by surface tension, i.e. the motion
of a grain at the free surface would increase the local curvature of
the liquid-air interface, which would decrease the local liquid
pressure and apply a suction that resists grain motion. In this
way, the surface grains have an apparent cohesion and surface
tension stabilises the free surfaces against gravity and prevents
slumping. To simulate the apparent cohesion, the rigid walls
were deleted and the surface grains were assigned partial
cohesion using parallel bonds in PFC2D that create a movable
membrane boundary that constrains the packing of grains similar
to reference [77,78]. These parallel-bonded surface grains are
shaded red and linked by red lines in Fig. 1eeg. With this
approach, a constraint force, FN , would be installed on any
membrane grain if the grain started to displace outward, and the
direction of FN for the membrane circle was determined from the
position of its two adjacent membrane grains (Fig. 1h and i). The
sample generation stage finished when the maximum magnitude
of grain velocity in the system was less than 107 m/s, and the
average contact force magnitude excluding membrane circles, F0 ,
was extracted to obtain the extent of the pre-existing force
network for each sample. The initial 2D packing fraction, f 0S:pk ,
was measured for each assembly by randomly distributing 5000
measurement circles with diameter 20d across the whole as-
sembly (see SI Section 7 for the descriptions of measurement
circle). To simulate the push-plate, a new rigid wall was intro-
duced as illustrated in Fig. 1e. The use of measurement circles as
a representative volume (RV) follows measurement approaches
in the DEM literature [79e81]. The concept of a circular RV in
DEM analysis here is different from the use of square/rectangular
REVs for image analysis suggested by Ref. [23].
The key parameters used for the DEM simulations are sum-
marised in Table 3. The solid phase densities differed in the four
virtual samples because experiments were either carried out at
different temperatures or with different bulk alloy compositions to
achieve the range of solid fractions, and the density was calculated
using the composition and temperature-dependent equations of
ref. [69]. However, the simulated results would not be significantly
altered if a single density had been used across all simulations. A
low sliding friction coefficient, m, was used at contacts as the liquid
phase is expected to provide lubrication, while the rolling-
212 T.C. Su et al. / Acta Materialia 163 (2019) 208e225

Fig. 1. The procedure to simulate semi-solid deformation adapting microstructures from deformation experiments: (a) in-situ synchrotron radiography imaging on 66% solid
sample before deformation, (b) 120 particles selected for shape analysis, (c) circled particle in (b) fitted using an elliptical equation, and (d) the corresponding two-circle grain used
to represent the shape of the fitted ellipse in (c). After adapting particle size and shape distribution by measuring 120 particles, a grain assembly (simulation C) was created which
was constrained by the push plate and membrane as shown in (e), and (f) is the deformation microstructure during push-plate displacement with rate duy =dt ¼ 30 mm s1 . The
black lines in (f) indicate the contact force chain and the thickness of each line is proportional to the contact force magnitude. The blue solid square in (eef) indicates the cor-
responding field of view in X-ray imaging. (geh) The cropped region from the dash-line square in (e) and (f), respectively. (h) Showing that constraint force normal to the membrane
was applied as the membrane boundary deformed, and (i) the determination of normal force direction from two adjacent circles. (For interpretation of the references to colour in
this figure legend, the reader is referred to the Web version of this article.)

resistance coefficient, mr , was specified from the consideration of
the non-spherical shape of particles and the existence of contact
surfaces rather than point contacts [82]. High bonding strengths
were assigned to the membrane contacts to prevent bond breakage.
Past DEM studies have shown that, provided changes in the
mean grain size do not significantly alter the number of grains in
the sample, many of the deformation features scale with the mean
grain size [83,84]. The contraction or dilation behaviour is sensitive
to initial solid fraction and the average grain size plays a more
minor role, mainly affecting the magnitude of volume change.
To account for the plastic deformation of grains during shear
deformation, the contact force-overlap relationship in Fig. 2a was
applied. This represents the flow curve of Al alloys deformed at
constant shear rate without work hardening after yielding ðUy ; Fy Þ
at homologous temperature near to 1.0 [85]. The setting of an
elastic unloading/reloading path between an unloading point ðUun ;
Fun Þ and residual overlap point ðUres ; 0Þ shown in Fig. 2a enables the
grain-grain contact to follow the behaviour of cyclic loading in al-
loys [86]. This plastic force-overlap relationship did not apply on
the grain-wall contacts or any bonded contacts between two cir-
cular membrane grains. The membrane boundary and plastic force-
overlap relationship were incorporated using the embedded
scripting language FISH in PFC2D, and the detailed code de-
scriptions are in SI Section 4.
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225 213

Table 3
Parameters used during shear deformation stage. The initial void ratio e0 ¼ ð1 
f 0S:pk Þ=f 0S:pk .

Property Unit Value

Grain density, r kg m3

Simulation A, for 44% solid sample 2643
Simulation B, for 50% solid sample 2658
Simulation C, for 66% solid sample 2620
Simulation D, for 71% solid sample 2625
Elastic normal stiffness, kn;e N m1 4
103
Slope of plastic flow regimea, kn;p N m1 1
107
Grain-wall normal stiffness, kn;gw N m1 6
103
Membrane circle normal stiffness, kn;mem N m1 0.013
Membrane bonding normal stiffness, kn;bond N m2 13
Tensile strength of bonding, sbond N m1 2
1010
Shear strength of bonding, tbond N m1 1
1010
Membrane bonding radius multiplier, lbond e 0.5
Normal-to-shear stiffness ratio, k e 2.0
Average contact force magnitude prior to def., F0 N
Simulation A, for 44% solid sample 0
Simulation B, for 50% solid sample 2.52
109
Simulation C, for 66% solid sample 1.64
108
Simulation D, for 71% solid sample 3.24
108
Initial 2D packing fraction:void ratio, f 0S:pk :e0 e
Simulation A, for 44% solid sample 0.78:0.28
Simulation B, for 50% solid sample 0.81:0.23
Simulation C, for 66% solid sample 0.89:0.12
Simulation D, for 71% solid sample 0.92:0.09
Total number of grains, N DEM e
grain
Simulation A, for 44% solid sample 7907
Simulation B, for 50% solid sample 7054
Simulation C, for 66% solid sample 5224
Simulation D, for 71% solid sample 8328
Minimum:maximum grain size, dDEM DEM
min :dmax
mm
Simulation A, for 44% solid sample 64:105
Simulation B, for 50% solid sample 91:165
Simulation C, for 66% solid sample 110:164
Simulation D, for 71% solid sample 90:160
Effective mean overlap prior to deformation, U0 m F0 =kn;e
Yielding contact overlap, Uy m U0 þ4.0
107
Yielding force, Fy N kn;e ,Uy
Normal force on membrane, FN N 0.25,Fy
Friction coefficient, m e 0.05 Fig. 2. (a) Normal force-overlap relationship used in the DEM simulation where F0 is
Rolling friction coefficient, mr e 0.5 the mean contact force prior to deformation related to initial solid fraction g0S , and U0 is
Simulation timestepb, Dt s 6.66
106 the corresponding mean contact overlap. (b) Installing the Al-Cu liquid domain on the
a
To ensure the mechanical stability of deformation simulation, there is a nonzero DEM grain assembly (Fig. 1e) by setting liquid lattice nodes and liquid boundary
slope of force-overlap relationship after yielding point corresponding to the red line conditions. The Al-Cu liquid domain is controlled by fixed bounce-back boundary
in Fig. 2a. (black dashed line), moving bounce-back boundary (green dashed line), and solid zone
b
The timestep setting allows the push plate have a constant vertical displacement outside of the membrane (grey-shaded area) corresponding to the lattice solid fraction
2:00
1010 m between two calculation cycles for all simulation datasets. ε ¼ 0. (c) The magnified view of the dashed grey box in (b) showing the DEM grains,
membrane circles, solid zone, and lattice nodes with spacing 40mm. (For interpretation
of the references to colour in this figure legend, the reader is referred to the Web
version of this article.)
2.5. LBM-DEM two-phase coupling

In order to incorporate the liquid phase, the lattice Boltzmann
method (LBM) was applied and LBM-DEM coupling as proposed by
Cook et al. [57] was used. The LBM solves the Navier-Stokes equa-
tions for the liquid on a regular grid. Each grid node contains a
packet of liquid particles that are allowed to move in horizontal,
vertical and diagonal directions. The grid nodes in the vicinity of a
liquid-solid surface act as a no-slip bounce-back boundary to the
interstitial liquid. The detailed description of governing equations
and key parameters in the LBM-DEM coupling adopted in this
research is given in SI Section 5.
The rectangular LBM grid extended beyond the DEM sample
boundaries (Fig. 2b). The lattice spacing was 40 mm, and the average
grain size d to lattice spacing ratios were 2e3; an immersed
boundary scheme was used to achieve sub-lattice resolution. The
liquid density, rL , calculated from Ref. [69], dynamic viscosity, mL ,
from Ref. [87], and LB liquid calculation timestep are listed in
Table 4. The kinematic viscosities (mL /rL ) used in the four simula-
tions were in the narrow range 6.10e6.75
107 m2/s, showing
that the Cu concentration in the liquid has a minor effect on the
liquid flow behaviour. Following past work on interstitial liquid
flow and permeability in 2D simulations of packed granular ma-
terials and porous media [88e90], the concept of a “hydrodynamic
radius” was used and set to be 0.8. This enables the liquid and solid
phases to both exist as percolating networks as in the thin sample
experiments. Before deformation, the liquid velocity was initialised
as zero. The LB bounce-back boundary condition [91] was applied
on boundary nodes (dashed black line in Fig. 2b), and the LB nodes
outside of the DEM membrane boundary were identified as an
effective “solid zone” to inhibit liquid Al from flowing out of the
DEM sample. The solid zone applied in the LBM simulations has no
effect on the DEM simulation of the grains. When the membrane
deforms, the LBM solid zone adapts to the new shape of the
membrane.
214 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
Table 4
Liquid parameters used in LBM-DEM coupling simulation.
Property Unit Value
Liquid density, rL kg m3
Simulation A, for 44% solid sample 2903
Simulation B, for 50% solid sample 2948
Simulation C, for 66% solid sample 2787
Simulation D, for 71% solid sample 2810
Liquid dynamic viscosity, m Pa s
Simulation A, for 44% solid sample 1.92
103
Simulation B, for 50% solid sample 1.99
103
Simulation C, for 66% solid sample 1.70
103
Simulation D, for 71% solid sample 1.75
103
LBM Simulation timestepa, dt s 6.66
104
a
The timestep setting in LBM was 100 times higher than DEM timestep listed in
Table 3.
During shear deformation, the push plate acts as a moving
bounce-back boundary [92] and nodes overlapping with the push
plate are automatically identified as solid. The liquid pressure
change Dp in response to shear deformation was derived from the
change of liquid density for each LBM node, and the speed of sound
cs , in the LBM domain (detail in SI Section 5):
Dpðxi ; tÞ ¼ ðrðxi ; tÞ  r0 Þc2s (10)
where r0 is the initial liquid density listed in Table 4, and rðxi ; tÞ is
the liquid density at position xi at simulation time t. The liquid drag
force and moment acting on DEM grains were updated by a cycle of
the LBM-DEM coupling algorithm (see SI Section 5 for detail).
3. Results and discussion
3.1. Analysis of radiography image sequences
Fig. 3aed shows processed in-situ X-ray images from four
datasets prior to deformation. Each image shows the solid fraction
in the beam path, L0S =L0alloy , at each pixel, calculated from the
transmitted intensities using Eq. (2) and SI-Eq. (3). In this conver-
sion, white represents solid (L0S =L0alloy ¼ 1) and black represents
liquid (L0S =L0alloy ¼ 0). Pixels with L0S =L0alloy value not falling in the
range [0,1] are coloured red, and are mainly regions of oxide and
pores. Fig. 3eeh are the LnS =Lnalloy field after a push-plate displace-
ment of 5d for each dataset. The changes in intensity of the LnS =Lnalloy
shading indicate changes in the local solid fraction due to grain
rearrangement and liquid flow.
The L0S =L0alloy fields were volume averaged using a REV of
pffiffiffiffiffiffi pffiffiffiffiffiffi
10d
10d to obtain the solid fraction field prior to deforma-
tion (Fig. 3iel) where the red pixels in Fig. 3aed were excluded
from the averaging process. Fig. 3iel shows that the solid phase
distribution before deformation is slightly segregated with a higher
solid fraction nearer the top of the FOV. This is especially evident for
the 44% solid dataset and is due to grain buoyancy in Al-15Cu
where the density of the solid is lower than the liquid [69]. After
a 5d increment of push-plate displacement, Fig. 3mep shows the
development of more heterogeneous solid fraction field. In all
samples, Fig. 3mep shows accumulation/compaction of grains and
expulsion of liquid (hotter colours) in front of the push plate where
the local g nS:REV is increased to z0:8. Additionally, for the 66% and
71% solid samples, a reduced solid fraction and increased liquid
fraction (cooler colours) with g nS:REV from 0.4 to 0.5 is present ahead
of the compaction zone.
The local changes in solid fraction correspond to volumetric
strains in the grain assembly, which are plotted as volumetric strain
maps εnvol in Fig. 3qet using Eq. (9) on the volume-averaged solid
fraction maps. It is clear from Fig. 3qet that, for initial solid frac-
tions of 44% and 50%, deformation is mostly contractive (blue) with
only small pockets of local dilation (red) and, as the initial solid
fraction increases, an increasing proportion of the sample un-
dergoes shear-induced dilation (red) and the magnitude of local
dilation increases. In Fig. 3q and r, the push-plate displacement on
the low solid fraction datasets (44% and 50%) developed a
compaction field with εnvol z  0:3, while the compaction zone in
66% and 71% solid datasets is less significant (Fig. 3s and t), where
εnvol falls in the range 0.2 ~ 0.1. It should also be noted that the
negative εnvol regions to the right of the FOV in Fig. 3s and the right-
bottom of the FOV in Fig. 3t represent local contraction corre-
sponding to the suction of liquid away to the dilating regions
elsewhere in the sample.
While there are differences between bulk 3D deformation and
the current study on thin sample deformation with interactions
with the confining walls, we note that the key features measured
here have also been measured in bulk 3D tomographic imaging [31]
and in microstructures after semi-solid deformation experiments.
For example, shear-induced contraction in samples containing a
loose solid network [17], shear-induced dilatation in samples con-
taining a dense solid network [9,19], and the localisation of dilat-
ancy [9].
The main deformation mechanisms that increase and decrease
the solid fraction and cause volumetric strain in Fig. 3qet, and the
resultant heterogeneous solid fraction field in Fig. 3mep are shown
in Fig. 4 using the 66% solid sample as an example. In region A, the
increase in solid fraction (contraction) can be seen to be due to
grains being pushed closer together and expelling some interstitial
liquid and, in this case, also due to compression of the individual
grains. In region B, the decrease in solid fraction (dilation) is due to
grains moving apart and liquid being drawn in interstitial. In region
B, all grains are in mechanical contact with their neighbours and,
therefore, the shear-induced dilation (dilatancy) is caused by grains
pushing one another apart under the action of compressive and
shear contact forces.
2D displacement fields were obtained by tracking the motion of
grains using DIC. Fig. 5aed shows the incremental displacement
vectors for a push-plate displacement of 0 to 5d for each dataset,
and the background image is the time averaged radiographs from
0 to 5d push-plate displacement for comparison. The vectors in
Fig. 5aed represent grain movement and show that, for the same
push-plate displacement, grains are displaced deeper into the
sample as the initial solid fraction increases. The displacement
vector field in DIC can be directly converted to shear strain (εxy )
fields as shown in Fig. 5eeh, where the background for each shear
strain field is selected as the image at 5d push-plate displacement.
In Fig. 5eeh, it is clear that the εxy is highly localised above the top-
right corner of push plate. The red region near to the parting plane
in the 44% solid sample (Fig. 5e) is relatively small, while the red
bands of localised shear strain form in 50% solid (Fig. 5f). A more-
developed region of shear strain along the parting plane can be
found in Fig. 5g and h which is wider and deeper than in Fig. 5eef.
From Fig. 5eeh, it can be seen that at the same push-plate
displacement a higher solid fraction induces a larger area of shear
strain (red) around the parting plane. The formation of blue regions
in Fig. 5g and h is due to εxy < 0 development near to the left parting
plane aligning with the top-left corner of push-plate (outside of the
left edge of the FOV).
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225 215

Fig. 3. Analysis of experimental X-ray datasets. (aed) Calculation of L0S =L0alloy (the fraction of solid at a beam path) field on synchrotron X-ray imaging before deformation and (eeh)
LnS =Lnalloy field on deformation after 5d increment of push-plate displacement. The conversion from L0S =L0alloy or LnS =Lnalloy field to image is shown in SI-Eq. 3. (iel) Average filtered
pffiffiffiffiffiffi pffiffiffiffiffiffi
results from (aed) with window size 10d
10d showing initial solid fraction field g 0S:REV . (mep) Average filtered results from (eeh) showing solid fraction evolution g nS:REV
during deformation. (qet) The evaluation of volumetric strain εnvol using Eq. (9). Images in the leftmost column correspond to the 44% solid sample, followed by 50%, 66%, and 71%
solid samples, respectively. White-line scale bars read 2mm.

3.2. Numerical simulation results and comparative study four experiments while varying few parameters: the initial packing
fraction (i.e. the solid fraction), the initial average contact stress,
It was found that a single set of optimised LBM-DEM parameters solid/liquid densities, and liquid viscosity directly linking to cor-
could reasonably reproduce the displacement and strain fields in all responding experiments. Note that, while the solid/liquid densities
216 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
Fig. 4. Microstructure of deformation experiment at 66% solid. Two regions from the image were selected for detailed observation in order to gain insights on strain localisation.
Region A shows high compaction along the y-direction, while the shear-induced dilation phenomenon can be clearly seen in region B when push-plate displacement ðuy Þ equals to
10d. Similar dilation behaviour can be observed in region B in the LBM-DEM simulation.
and liquid viscosity were calculated for the relevant temperature
and phase compositions, the variation in these parameters is small
and the simulated results would not be significantly altered if a
single density and liquid viscosity value had been used across all
simulations. Thus, effectively two parameters were varied. The
optimised set of parameters including contact stiffness values and
friction coefficients is shown in Table 3, and was found by an iter-
ative approach to reproduce the deformation microstructure in
each experiment across the full range of applied displacements.
This was done by starting with literature values for stiffness [64,93]
and friction coefficient [64,82,93,94], and then varying these to
obtain a good correlation (enorm in Eq. (11) within 10%) of the 1D
displacement profile between simulation and experiment.
To enable direct comparison DIC analysis was performed on
DEM simulated data, using the same DIC techniques and parame-
ters as in the experimental analysis. To do this, DEM grains were
coloured by random greyscale to construct a speckle pattern, the
pixel size and FOV of DEM microstructure plots were adjusted to be
the same as in radiography images, and the same DIC parameters
were used (Table 2). Fig. 5iel shows the result for the displacement
fields, and a good correlation of vector fields between experiment
(Fig. 5aed) and simulation (Fig. 5iel) can be seen. It can also be seen
that the DIC shear strain fields from the simulations (Fig. 5mep)
correctly capture many of the shear strain localisation character-
istics in the experiments (Fig. 5eeh). For example, the shape and
extent of the red localised shear zones, and the influence of initial
solid fraction on the shear strain fields are well reproduced. Also
note from Fig. 5g and h, that the shear strain is positive just in front
of the push plate for the 66% solid sample while it is negative for the
71% solid sample. This is because the boundary conditions are
slightly different for these samples. In the 66% solid sample
(Fig. 5g), the sample sides did not touch the Al2O3 side walls and, in
the 71% solid sample, the left edge of the sample is in contact with
the Al2O3 side wall (as introduced in section 2.4). The simulations in
Fig. 5o and p used the boundary conditions measured for each
experiment and gave a similar result. This is because constraint
from the left side wall prevents grain displacement to the left,
resulting in a change in shear strain direction.
The displacement vector fields in Fig. 5iel were then interpo-
lated and scanned to construct grain displacement profiles as
shown in Fig. 6a, which plots the grain displacement relative to the
push-plate displacement at various distances ahead of the push-
plate along a 1D line starting at the middle of the push-plate as
indicated by the vertical dashed line in Fig. 5m. Comparison of the
1D displacement profiles between experiments and simulations
after a 5d increment of push-plate displacement confirms that the
simulation reproduces the gradient of the displacement as well as
the higher relative displacement in higher solid fraction datasets.
To quantify this, Fig. 6b is a bar chart of the normalised error of
simulated displacement profiles compared with the experiments,
enorm , with increments of push-plate displacement from 1d to 5d.
The normalised error is defined as:
1 XY2
dEXP ðYÞ dDEM ðYÞ
enorm ¼ 
100% (11)
Y2  Y1 Y¼Y uEXP
y uDEM
y
1
where Y1 ¼ 250mm and Y2 ¼ 3000mm are the nearest and farthest
points from the middle point of the push plate in the FOV, uEXP
y and
uDEM
y are the push-plate displacement lengths, and dEXP ðYÞ and
dDEM ðYÞ are the interpolated displacement magnitude in scan point
Y in an experiment and simulation. The scan resolution is 1 mm.
From Fig. 6b, it can be seen that the error value enorm is <10% for all
experiment/simulation pairs, indicating good agreement between
simulation and experiment.
We next examine how closely the simulations can reproduce the
heterogeneous strain patterns observed in the radiography data-
sets. Comparison between localised strains can be simplified by the
conversion from two-dimensional strain fields to averaged one-
dimensional line profiles by averaging along x or y within a ROI,
as shown from Fig. 7 to Fig. 10. In (c) and (d) of Figs. 7e10, the red
lines correspond to LBM-DEM simulations and the black lines
correspond to experimental measurements taken from the red ROI
(box outlined by the red-dashed line in (a) and (b) of Figs. 7e10,
which is averaged along y). The red ROI was selected to investigate
the development of strain heterogeneity across the parting plane.
In (e) and (f) of Figs. 7e10, the blue lines correspond to LBM-DEM
simulations and the black lines correspond to experimental mea-
surements taken from the blue ROI ahead of the push plate (box
outlined by blue-dashed line in (a) and (b) of Figs. 7e10, which is
averaged along x). The blue ROI was selected to investigate the
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225 217

Fig. 5. (aed) The superimposition of time-averaged raw images and displacement vectors for experiments. Each time-averaged image is the averaging of radiographic image
sequences from the onset of deformation to the deformed image after 5d increment of push-plate displacement, and the displacement vectors are derived from digital image
correlation (DIC). The spacing between two adjacent displacement vectors is equal to the DIC subset size of that dataset. (eeh) Shear strain εxy field for experiments derived from
DIC. The background images were chosen as the push-plate displacement ¼ 5d radiographs. (iel) The superimposition of time-averaged speckled DEM grain plots and displacement
vectors. (mep) Shear strain εxy field for simulations derived from DIC. The dashed line in (m) indicates the middle of the push-plate. The first row corresponds the 44% solid sample
and simulation A, followed by 50% solid/simulation B, 66% solid/simulation C, and 71% solid/simulation D. Black-line scale bars read 2mm.

transitions of contraction-to-dilation response ahead of the push
plate. The strain profiles acquired from triangulation are “jumpy”
because the strains are calculated from the relative position
changes of groups of three adjacent grains without any averaging
by a REV.
It can be seen in Figs. 7e10 that, in general, there is good
agreement between experiments and simulations. Specifically, the
strain profiles from the 44% solid fraction experiment and simula-
tion A (Fig. 7) show net-contraction in front of the push plate from
the εnvol profile in X-ray intensity processing on the experiment and
divergence (εxx þ εyy ) profiles by DIC and triangulation on
simulation A in Fig. 7c and e, and the distinctively negative ( < 
0:1) volumetric strain or divergence to yz12d in the blue ROI
(Fig. 7e) and at x < 0 in the red ROI (Fig. 7c) is captured well. The
average shear strain profiles show minimal strain localisation in
both the experiment and simulation (Fig. 7e and f).
For deformation at a slightly higher initial solid fraction (50%
solid sample and simulation B) shown in Fig. 8, the volumetric
strain εnvol in the experiment and divergence profiles in the simu-
lation are contractive in Fig. 8c and e, but less negative than in
Fig. 7c and e. The shear strain profiles (Fig. 8d) go higher than in the
44% solid/simulation A of Fig. 7, indicating more strain-localisation
218 T.C. Su et al. / Acta Materialia 163 (2019) 208e225

Fig. 6. (a) Line plot showing relative displacement (grain displacement magnitude
divided by push-plate displacement) profiles of shear experiment and simulation
deformation microstructures from a scan of displacement length field after 5d incre-
ment of push-plate displacement. (b) Bar chart showing normalised error values enorm
(defined in Eq. (11)) with different push-plate displacement lengths for each dataset.
at higher solid fraction. Shear strain development occurs ahead of
the push plate to yz15d in both the experiment and corresponding
simulation B, as shown in Fig. 8f.
Fig. 9 shows that deformation at higher initial solid fraction
(66%) gives rise to localisation of both positive εnvol and shear strain
εxy in a similar location (xz  5  3d in the red ROI of Fig. 9c and
d), which indicates a dilatant region [26]. The broad positive
(dilative) εnvol and divergence zone to xz12d in Fig. 9c is a distinc-
tive difference of strain response compared with the contractive
low solid fraction datasets in Figs. 7c and 8c. In Fig. 9d, peaks of the
shear strain profile are approximately located at the parting plane
(xz0). The co-location of the positive zone in divergence/volu-
metric strain profiles and εxy peaks in Fig. 9c and d gives a dilatant
shear band width at 66% solid within the range 10d to 15d, which is
consistent with the measurements in past bulk rheology tests
[9,17,95]. Analysis of the blue ROI, captures the contraction field
immediately ahead of push-plate (Fig. 9e), followed by relatively
smooth and positive profiles (normalised y axis from 2d to 21d), in
both the experiment and corresponding simulation C. The
contraction field at the push plate front is due to local compaction
at this location. Fig. 9c captures another negative (contractive)
volumetric strain region to the far-right of the red ROI starting at
x ¼ 12d, both in the εnvol profile of the experiment and the diver-
gence profiles from LBM-DEM simulation B, due to liquid drainage
at this location (i.e. the suction of liquid from the far-right of the
FOV to feed dilation near the parting-plane).
Fig. 10 is the response of the highest solid fraction dataset
studied here (71%) which had the most net-dilative response and
simulation D. Fig. 10c shows a liquid-enriched dilation zone at
x < 7d, followed by a contraction zone to the right of the red ROI
(negative εnvol or divergence at x > 10d) both in the experiment and
simulation D strain profiles. In the red ROI, both the experimental

Fig. 7. Direct comparison among strain profiles from 44% solid sample and simulation A after 5d increment of push-plate displacement. (a) Two regions of interest (ROIs) to study
the profiles of strain. The first ROI outlined by the red-dashed line contains the region in front of push plate ðx < 0Þ, the parting plane ðx ¼ 0Þ, and right-hand side of push plate
ðx > 0Þ. The second ROI outlined by the blue-dashed line contains the region in front of the push plate ðy ¼ 0Þ to y ¼ 25d (b) ROIs for the strain profile from simulation deformation
microstructure in which the DEM grains were coloured semi-transparent light grey and the liquid domain were coloured dark grey. (c) The corresponding volumetric strain in (a)
and divergence in (b) using the red ROI. The x-direction is scaled by the mean grain size d and the parting plane is set as the origin ðx ¼ 0Þ. XIP ¼ X-ray Intensity Processing. (d) The
averaged shear strain profiles for the red ROI in (a) and (b). (e) The volumetric strain or divergence profiles for the blue ROI. The origin corresponds to the bottom edge of blue ROI.
(f) The shear strain profiles for the blue ROI. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225 219

Fig. 8. Direct comparison among strain profiles from 50% solid sample and simulation B after 5d increment of push-plate displacement. (a) Two regions of interest (ROIs) to study
the profiles of strain. (b) ROIs for the strain profile from the simulation deformation microstructure. (c) The corresponding volumetric strain in (a) and divergence in (b) using the
red ROI. (d) The averaged shear strain profiles for red ROI in (a) and (b). (e) The volumetric strain or divergence profiles for the blue ROI. (f) The shear strain profiles for the blue ROI.
(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Fig. 9. Direct comparison among strain profiles from 66% solid sample and simulation C after 5d increment of push-plate displacement. (a) Two regions of interest (ROIs) to study
the profiles of strain. (b) ROIs for the strain profile from the simulation deformation microstructure. (c) The corresponding volumetric strain in (a) and divergence in (b) using the
red ROI. (d) The averaged shear strain profiles for the red ROI in (a) and (b). (e) The volumetric strain or divergence profiles for the blue ROI. (f) The shear strain profiles for the blue
ROI. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
220 T.C. Su et al. / Acta Materialia 163 (2019) 208e225

Fig. 10. Direct comparison among strain profiles from 71% solid sample and simulation D after 5d increment of push-plate displacement. (a) Two regions of interest (ROIs) to study
the profiles of strain. (b) ROIs for the strain profile from the simulation deformation microstructure. (c) The corresponding volumetric strain in (a) and divergence in (b) using the
red ROI. (d) The averaged shear strain profiles for the red ROI in (a) and (b). (e) The volumetric strain or divergence profiles for the blue ROI. (f) The shear strain profiles for the blue
ROI. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

εnvol profile and simulated εxx þ εyy profile by triangulation show

more positive dilative strains ahead of the push plate (x < 0d) than
in Fig. 9c, but the peak width of the εxy profile in Fig. 10d is similar to
Fig. 9d (from 10d to 15d). The blue ROI contains a very small
contractive region immediately ahead of the push plate, both in the
71% solid experiment and simulation D (Fig. 10e), and the εnvol profile
in the experiment then seems to have a periodic pattern with
multiple peaks and valleys. The explanation of this phenomenon is
incomplete liquid feeding during deformation at high solid fraction.
As a result, the εnvol field of the 71% solid sample (Fig. 3t) shows some
island-like contractive areas where liquid is sucked to adjacent
dilating zones. Those locally dilative zones correspond to peaks at
yz10:5d and 17:5d in Fig. 10e. The complex heterogeneous liquid
flow phenomena were not completely captured by the LBM-DEM
simulation, but a reasonable correlation with experiment in
Fig. 10e can still be observed.
While the simulations are 2D, the thin-sample experiments can
be considered to be 2.5D with interactions from the confining walls.
For the contractive region ahead of the push plate, the LBM-DEM
simulation captures grain deformation by overlap between DEM
grains which is tuned in the simple approach to plasticity in Fig. 2a
to match the overlap in the through-thickness averaged radio-
graphs. Therefore, although the 2D simulation model does not
capture the change in thickness directly, it does capture the
‘effective grain overlap’ in regions of thickening due to compaction.
For other regions, the change of sample thickness was low with
overlap between grains of <0.1 d both in experiments and simu-
lations, where the assumption of in-plane deformation is
reasonable. Fig. 11. Examples of the correlation between liquid behaviour in experiments and sim-
ulations. (a) Liquid expelled out at  11d push-plate displacement in low solid fraction
Comparison of all experimental and simulated profiles in alloy (44%), (b) increased liquid pressure field in simulation A, (c) meniscus sucked in
Figs. 7e10 demonstrates that the coupled LBM-DEM simulations phenomenon at  28d push-plate displacement in high solid fraction alloy, and (d)
quantitatively captured many of the key features of the decreased liquid pressure field at the right edge of the push plate in simulation C.
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
heterogeneous volumetric and shear strain fields at solid fractions
in the range 44e71 vol% solid, by only altering the initial packing
fraction and initial average contact stress in the simulations. An
important feature of the LBM-DEM simulations is that the local
changes in packing density arise naturally as emergent phenomena.
For example, in Fig. 4, Region B undergoes shear-induced dilation
due to grains pushing each other apart in response to contact forces
in both the experiment and simulation. The LBM-DEM simulations
also reproduce further important phenomena observed in the ex-
periments. For example, Fig. 11a shows that as the push plate
continued to 11d displacement in the lowest solid fraction sample
(44%), liquid was expelled from the sample-air interface at the top-
left of the sample, indicating that an excess liquid pressure devel-
oped in this region. The simulated liquid pressure field in Fig. 11b
correctly predicts a positive (red) change in liquid pressure in this
region at 11d displacement and agrees with the location of the
liquid expulsion. In Fig. 11b, note that the liquid pressure is highest
on the lower-left-side of the Figure. However, the lower-left-side of
the Figure is not a free surface; it is the edge of the experimental
FOV. In SI Section 9, the full simulation domain is shown (which
equals the whole experimental sample), where it can be seen that
the true left-side free surface of the sample has a liquid pressure
that is lower than the liquid pressure at the top of the FOV in
Fig. 11b. It also shows that the region where liquid is expelled in
Fig. 11a is the region with the highest liquid pressure near a surface
in Fig. 11b.
On the other hand, Fig. 11c shows that, in the highest solid
fraction sample, menisci are sucked in to the liquid at the free
surface near the right edge of the push plate during deformation.
Fig. 12. Plots of liquid velocity magnitude field (normalised by push-plate velocity
magnitude) surrounding grains for (a) simulation A (b) B, (c) C, and (d) D after 5d
increment of push-plate displacement. The black solid-outline square indicates the
corresponding FOV in radiography imaging in deformation experiments, and the
radius of each grain is reduced by 80% of its original size to show the effect of hy-
drodynamic radius on the LBM liquid domain. The left-hand side wall installed in
simulation D is consistent with the radiograph shown in Fig. SI-6d.
221
This is in agreement with the simulated liquid pressure field in
Fig. 11d where a decreased liquid pressure (blue) region exists at
this location. The different liquid pressure behaviour reflects the
net-contractive response at low solid fraction (44%, Fig. 3q) and
net-dilative shear response at high solid fraction (71%, Fig. 3t), and
is a result of the coupling between local volumetric strains in the
grain assembly and the liquid pressure field. This behaviour is ex-
pected to depend on the permeability and the strain rate and it was
found that, when similar samples with initial solid fractions >73%
were deformed, shear cracking occurred (not shown here).
The simulated push plate penetrates the compacted grain as-
sembly in the DEM domain, and simultaneously acts as a moving
solid boundary on the LBM liquid domain. Fig. 12 illustrates the
simulated liquid flow field after a push plate displacement of 5d.
This field was constructed by colouring the LBM nodes by the
relative velocity magnitude normalised by the push plate
displacement rate. In Fig. 12, the localised high liquid flow rate in
response to the penetration is very clear among all deformation
datasets, but more prominent liquid flow across the whole sample
in Figs. 12c and d is observed. The liquid flow is induced by the drag
(momentum exchange) from adjacent DEM grains. Also in Fig. 12c
and d, there is a subtle liquid flow towards the specimen centre
from the membrane boundary with relative velocity magnitude
<0.3. This can be related to a pressure differential that draws liquid
to regions undergoing high shear-induced dilation. This is also seen
in the experiments as a local contraction (grains moving closer
together) to the right and bottom-right of Fig. 3s and t.
3.3. Stress-deformation response
A well-calibrated set of LBM-DEM simulation results can offer
extra information that is not available in X-ray imaging. Fig. 13a is
the evolution of normal push-plate stress tracked to 10d push-plate
displacement, using the sum of grain-plate normal contact forces
divided by the length of the push plate front similar to reference
[96]. The normal stress gradually increases in lower solid fraction
simulations A and B, while a peak push-plate stress can be found at
~2d displacement for higher solid fraction simulation C, and at 3d
displacement for the highest solid fraction simulation D. It is noted
that the predicted push-plate stress evolution cannot be compared
with deformation experiments since the friction between the push-
plate and Al2O3 window compromises our ability to accurately
measure the push-plate load. Still, both the trend of increased peak
stress in higher solid fraction simulations and the strain softening
phenomenon observed in simulation C and D are consistent with
previous deformation tests on granular/particulate materials
including semi-solid alloys [10,19,31,97], experimental soils
[96,98,99], and DEM-simulated granular soils [100,101]. Fig. 13b
shows the evolution of total volumetric strain, which was calcu-
lated from:
 
Adef
εDEM
vol ¼ ln (12)
A
where εDEM
vol
is the volumetric strain directly evaluated from the
change of area in the simulation samples, A and Adef are the area of
the initial and deformed sample, which were obtained from the
area of the polygon that joins the centroids of all surface bonding
grains minus the push-plate penetration area. Fig. 13b clearly
shows the transition from net-contractive simulation sets A and B
with low initial solid fraction, to net-dilative simulation sets C and
D with higher initial solid fraction. The reduction of sample area in
A and B occurs because the initially loosely packed solid network in
a low-solid fraction LBM-DEM simulation is prone to develop a
222 T.C. Su et al. / Acta Materialia 163 (2019) 208e225

Fig. 13. (a) Evolution of normal push-plate stress, (b) volumetric strain ε0vol , and (c) the mean of total liquid pressure change versus 10d push-plate displacement. (d) Schematic
illustration of the measurement regions for deriving local parameters. Diameter of measurement circles is 20d in order to obtain representative stresses, and there is a 1:5d gap
between measurement circles and the push plate front. The horizontal position of the grey circle is in the middle of push plate, and the horizontal position of the blue circle is at the
top-right corner of the push plate. The corresponding (e-f) q  p0, (geh) e  uy =d, and (iej) e  log p0 plots are shown for black and blue measurement circles. Arrows in (iej) indicate
the initial state for each simulation. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

highly localised contractive field, resulting in the closure of liquid
interstices. Simulation C, on the contrary, deforms with slightly net
positive εDEM
vol
behaviour because the contact force can be trans-
mitted to the membrane to push the boundary outward and
compensate for the local contractive response ahead of the push
plate. Moreover, deformation in simulation D causes many of the
liquid interstices to expand during grain rearrangement outside of
the local compacted zone as well as the increased ability to push
the enclosing membrane away, resulting in the net dilative defor-
mation behaviour. Fig. 13b also shows the slight decrease of εDEM vol
after 6d penetration on simulation C and 7:5d penetration on
simulation D. This is because the total solid fraction of the samples
is decreased during net dilation, which increases the compress-
ibility of the whole assembly.
The complex liquid behaviour at the free surface of semi-solid
samples in the experiments can be related to the localised liquid
pressure change Dp as shown in Fig. 11. We can then investigate the
relationship between the change of volume of liquid-saturated
samples and the overall liquid pressure changes. To get a repre-
sentative trend, the mean liquid pressure change Dp was derived
across the whole DEM assembly:
PNnode
i¼1
Dpðxi ; tÞ
Dp ¼ DpðtÞ ¼ (13)
Nnode
where Nnode is the number of LBM nodes inside the DEM assembly.
Fig. 13c is the evolution of Dp for each simulation set and com-
parison with Fig. 13b shows the interplay between the global liquid
pressure change and the volumetric strain, The net-contractive
simulations A and B have a net increase of liquid pressure in
response to net contraction. The high solid fraction simulations C
and D have a slightly positive pressure change up to 1.5d defor-
mation, followed by a decrease in liquid pressure in response to the
global dilation, and the liquid pressure is near-constant once the
peak dilatational volumetric strain is reached (Fig. 13b).
It is useful to investigate the local stress evolution in a circular
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
RV by explicit tracking of contact force vectors in the DEM system
especially for the region in front of the push plate and the parting
plane. Fig. 13d illustrates the setting of measurement circles in the
DEM system to derive the local parameters such as the effective
mean stress p0 , deviatoric stress q, and void ratio e. In two-
dimensional DEM, the relationship between void ratio e and grain
packing fraction (i.e. solid fraction) fS:pk is e ¼ ð1  fS:pk Þ=fS:pk . These
local parameters were tracked to examine whether the DEM sys-
tem behaves as a granular material such as a sand [22]. The
diameter of the measurement region was chosen to be 20d and
includes ~250 grains; this is sufficiently large to obtain the effec-
tively averaged behaviour of the regions of interest. Fig. 13e and f
illustrate the evolution of the q versus p0 relationship to 10d
displacement. Before shear deformation, the effective stress p0 re-
flects the required mean stress on the DEM assembly to reach the
desired packing fraction, and the mean stress values are similar
between black and blue circles for every simulation set. The q- p0
curves all start from q ¼ 0 and the stresses are increasing for
simulation A and B. It is noted that the low solid fraction simulation
sets (red and blue in Fig. 13e and f) soon have an approximately
constant q= p0 ratio and follow the regression line q ¼ 0:428p0 ,
which can be related to the well-defined critical state in soil me-
chanics [22] stating that a frictional fluid should reach a macro-
scopic frictional constant M ¼ q= p0 . In contrast, the maximum
deviatoric stress occurs in simulations C and D at 0.98
104 N/m
and 2.35
104 N/m for the black circle, and 0.64
104 N/m and
2.07
104 N/m in the blue circle, respectively. After reaching the
maximum deviatoric stress, both p0 and q stresses decrease. The
simulation C then reaches the q ¼ 0:428p0 regression line, while the
q- p0 curves in simulation D show a nearly 50% decrease of
stresses from the peak deviatoric stress point and moves towards
the regression line (but does not reach it by 10d push-plate
displacement).
The evolution of void ratio is shown in Fig. 13g and h. The high
contraction field in front of the push plate developed at an early
stage of deformation (3d) reduces the void ratio by about 0.03 in
the black circle and about 0.01 in the blue circle. The void ratio
decrease corresponds to the increase of local mean effective
stresses as shown in Fig. 13i and j. There is a log-linear relationship
between the final void ratio and the final value of p0 ; this corre-
sponds to the critical state line concept accepted for granular soil
behaviour [22]). The increase of void ratio in high solid fraction
deformation is due to shear-induced dilation in both the black and
blue circles.
Whilst this research used globular grains to prevent significant
shape change during isothermal experiments and enable an effi-
cient model, casting processes often create more complex mor-
phologies such as equiaxed dendrites. Experiments [102] and DEM
studies [64] have shown that dendrites form a solid network at
lower solid fraction than globular grains due to the high liquid
fraction within dendrite envelopes and due to the more complex
shape of the envelopes. These factors will also alter the perme-
ability [103], give more contacts per grain, and affect the local
stresses acting on dendrites [64]. Future research can extend the
LBM-DEM approach to dendritic microstructures in the future.
4. Conclusions
Time-resolved synchrotron X-ray radiography and coupled
LBM-DEM simulations have been applied to investigate shear
deformation mechanisms in equiaxed globular AleCu alloys at
44e71% solid. Microstructural and macroscopic responses to load
have been quantified by intensity processing and digital image
223
correlation (DIC) for radiographic imaging with >1000 grains in the
field of view (FOV), and also reproduced by LBM-DEM after finding
a single set of DEM parameters calibrated across all deformation
experiments from 44 to 71% solid. From the results obtained by the
various analysis approaches, the following conclusions can be
drawn:
 The volumetric strain during semi-solid deformation due to
local changes of solid fraction is strongly influenced by the
initial solid fraction. A contractive strain field developed in shear
or compressive deformation in alloys initially containing 44%
and 50% solid. In contrast, significant net-dilation occurred
during shear on the 66% and 71% solid samples.
 The mechanism of solid fraction increase during deformation is
grains being pushed together by compaction, or by the liquid
being sucked away. The solid fraction decrease, on the other
hand, is due to shear-induced dilation by grains pushing one
another apart during grain rearrangement in a solid network,
similar to phenomena in compacted granular/particulate
materials.
 Deformation was highly heterogeneous and the strain was
strongly localised. Both the experiments and simulations agree
with the trend of net-contraction ahead of the push plate and, in
high solid fraction deformation, the co-location of dilative
volumetric strain and shear strain in a region with width of
10e15 mean grains wide. Additionally, a contraction field
formed outside the locally dilating regions in high solid fraction
samples due to liquid being sucked away to feed dilation.
 The coupled LBM-DEM model naturally simulates the change of
packing density in primary Al grains as an emergent phenom-
enon, and captures the complex liquid behaviour recorded in
radiographs from the simulated liquid pressure change.
 The calibrated LBM-DEM model provided information on both
the local stresses and void ratios; these data show that the
material exhibits a load deformation response that can be
described by the critical state framework for soil behaviour.
Acknowledgements
The in-situ observations were supported by a Grant-in-Aid for
Scientific Research (S) (No. 17H06155). Experiments were con-
ducted at the SPring-8 Synchrotron on BL20B2 under Proposal No.
2015A1318 and 2017B1523. Analysis was carried out under the
EPSRC Grant EP/K026763/1. T.C. Su gratefully acknowledges a
President's scholarship from Imperial College London.
Appendix A. Supplementary data
Supplementary data to this article can be found online at
https://doi.org/10.1016/j.actamat.2018.10.006.
References
[1] A. Kaye, A. Street, Die Casting Metallurgy, Butterworth Scientific, London,
1982.
[2] E.J. Vinarcik, High Integrity Die Casting Processes, John Wiley & Sons, New
Jersey, 2002.
[3] W. Andresen, Die Cast Engineering: a Hydraulic, Thermal, and Mechanical
Process, CRC Press, New York, 2004.
[4] X.S. Huang, L.J. He, G.B. Mi, P.J. Li, Characteristics of defect bands and their
formation mechanisms in A356 wheel fabricated by horizontal squeeze
casting, Mater. Sci. Technol. 31 (4) (2015) 400e408.
[5] M. Masoumi, H. Hu, Influence of applied pressure on microstructure and
tensile properties of squeeze cast magnesium Mg-Al-Ca alloy, Mater. Sci.
Eng., A 528 (10e11) (2011) 3589e3593.
[6] R. Cook, P.G. Grocock, P.M. Thomas, D.V. Edmonds, J.D. Hunt, Development of
the twin-roll casting process, J. Mater. Process. Technol. 55 (2) (1995) 76e84.
[7] C. Gras, M. Meredith, J.D. Hunt, Microdefects formation during the twin-roll
224 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
casting of Al-Mg-Mn aluminium alloys, J. Mater. Process. Technol. 167 (1)
(2005) 62e72.
[8] Z. Bian, I. Bayandorian, H.W. Zhang, G. Scamans, Z. Fan, Extremely fine and
uniform microstructure of magnesium AZ91D alloy sheets produced by melt
conditioned twin roll casting, Mater. Sci. Technol. 25 (5) (2009) 599e606.
[9] C.M. Gourlay, A.K. Dahle, Dilatant shear bands in solidifying metals, Nature
445 (7123) (2007) 70e73.
[10] C.M. Gourlay, B. Meylan, A.K. Dahle, Shear mechanisms at 0-50% solid during
equiaxed dendritic solidification of an AZ91 magnesium alloy, Acta Mater. 56
(14) (2008) 3403e3413.
[11] C.M. Gourlay, H.I. Laukli, A.K. Dahle, Defect band characteristics in Mg-Al and
Al-Si high-pressure die castings, Metall. Mater. Trans. 38A (8) (2007)
1833e1844.
[12] S. Otarawanna, C.M. Gourlay, H.I. Laukli, A.K. Dahle, The thickness of defect
bands in high-pressure die castings, Mater. Char. 60 (12) (2009) 1432e1441.
[13] M.C. Flemings, Solidification Processing, McGraw-Hill, New York, 1974.
[14] C.P. Chen, C.Y.A. Tsao, Semi-solid deformation of non-dendritic structures. 1.
Phenomenological behavior, Acta Mater. 45 (5) (1997) 1955e1968.
[15] S. Otarawanna, H.I. Laukli, C.M. Gourlay, A.K. Dahle, Feeding mechanisms in
high-pressure die castings, Metall. Mater. Trans. 41a (7) (2010) 1836e1846.
[16] M.S. Kim, S.H. Kim, H.W. Kim, Deformation-induced center segregation in
twin-roll cast high-Mg AleMg strips, Scripta Mater. 152 (2018) 69e73.
[17] B. Meylan, S. Terzi, C.M. Gourlay, A.K. Dahle, Dilatancy and rheology at 0-60%
solid during equiaxed solidification, Acta Mater. 59 (8) (2011) 3091e3101.
[18] E. Tzimas, A. Zavaliangos, Mechanical behavior of alloys with equiaxed
microstructure in the semisolid state at high solid content, Acta Mater. 47 (2)
(1999) 517e528.
[19] T. Sumitomo, D.H. StJohn, T. Steinberg, The shear behaviour of partially so-
lidified Al-Si-Cu alloys, Mater. Sci. Eng., A 289 (1e2) (2000) 18e29.
[20] O. Reynolds, LVII. On the dilatancy of media composed of rigid particles in
contact. With experimental illustrations, Philos. Mag. Series 5 20 (127)
(1885) 469e481.
[21] W. Powrie, Soil Mechanics: Concepts and Applications, second ed., Taylor &
Francis, New York, 2004.
[22] A. Schofield, P. Wroth, Critical State Soil Mechanics, McGraw-Hill, New York,
1968.
[23] C.M. Gourlay, A.K. Dahle, T. Nagira, N. Nakatsuka, K. Nogita, K. Uesugi,
H. Yasuda, Granular deformation mechanisms in semi-solid alloys, Acta
Mater. 59 (12) (2011) 4933e4943.
[24] C.M. Gourlay, C. O'Sullivan, J. Fonseca, L. Yuan, K.M. Kareh, T. Nagira,
H. Yasuda, Synchrotron radiography studies of shear-induced dilation in
semisolid Al alloys and steels, JOM 66 (8) (2014) 1415e1424.
[25] T. Nagira, C.M. Gourlay, A. Sugiyama, M. Uesugi, Y. Kanzawa, M. Yoshiya,
K. Uesugi, K. Umetani, H. Yasuda, Direct observation of deformation in semi-
solid carbon steel, Scripta Mater. 64 (12) (2011) 1129e1132.
[26] T. Nagira, H. Yokota, S. Morita, H. Yasuda, M. Yoshiya, C.M. Gourlay,
A. Sugiyama, K. Uesugi, K. Umetani, Characterization of shear deformation
based on in-situ observation of deformation in semi-solid Al-Cu alloys and
water-particle mixture, ISIJ Int. 53 (7) (2013) 1195e1201.
[27] J. Fonseca, C. O'Sullivan, T. Nagira, H. Yasuda, C.M. Gourlay, In situ study of
granular micromechanics in semi-solid carbon steels, Acta Mater. 61 (11)
(2013) 4169e4179.
[28] T. Nagira, S. Morita, H. Yokota, H. Yasuda, C.M. Gourlay, M. Yoshiya,
A. Sugiyama, K. Uesugi, A. Takeuchi, Y. Suzuki, In situ observation of defor-
mation in semi-solid Fe-C alloys at high shear rate, Metall. Mater. Trans. 45A
(12) (2014) 5613e5623.
[29] S. Zabler, A. Rack, A. Rueda, L. Helfen, F. Garcia-Moreno, J. Banhart, Direct
observation of particle flow in semi-solid alloys by synchrotron X-ray micro-
radioscopy, Phys. Status Solidi 207 (3) (2010) 718e723.
[30] S. Zabler, A. Ershov, A. Rack, F. Garcia-Moreno, T. Baumbach, J. Banhart,
Particle and liquid motion in semi-solid aluminium alloys: a quantitative in
situ microradioscopy study, Acta Mater. 61 (4) (2013) 1244e1253.
[31] K.M. Kareh, P.D. Lee, R.C. Atwood, T. Connolley, C.M. Gourlay, Revealing the
micromechanisms behind semi-solid metal deformation with time-resolved
X-ray tomography, Nat. Commun. 5 (2014) 1e7.
[32] B. Cai, P.D. Lee, S. Karagadde, T.J. Marrow, T. Connolley, Time-resolved syn-
chrotron tomographic quantification of deformation during indentation of
an equiaxed semi-solid granular alloy, Acta Mater. 105 (2016) 338e346.
[33] S. Terzi, L. Salvo, M. Suery, N. Limodin, J. Adrien, E. Maire, Y. Pannier,
M. Bornert, D. Bernard, M. Felberbaum, M. Rappaz, E. Bollerr, In situ X-ray
tomography observation of inhomogeneous deformation in semi-solid
aluminium alloys, Scripta Mater. 61 (5) (2009) 449e452.
[34] K.M. Kareh, C. O'Sullivan, T. Nagira, H. Yasuda, C.M. Gourlay, Dilatancy in
semi-solid steels at high solid fraction, Acta Mater. 125 (2017) 187e195.
[35] A. Zavaliangos, Modeling of the mechanical behavior of semisolid metallic
alloys at high volume fractions of solid, Int. J. Mech. Sci. 40 (10) (1998)
1029e1041.
[36] J.J. Wang, A.B. Phillion, G.M. Lu, Development of a visco-plastic constitutive
modeling for thixoforming of AA6061 in semi-solid state, J. Alloy. Comp. 609
(2014) 290e295.
[37] X. Hu, Q. Zhu, H. Atkinson, H. Lu, F. Zhang, H. Dong, Y. Kang, A time-
dependent power law viscosity model and its application in modelling semi-
solid die casting of 319s alloy, Acta Mater. 124 (2017) 410e420.
[38] M.H. Sheikh-Ansari, M. Aghaie-Khafri, Shear localization in semi-solid
deformation: a bifurcation theory approach, Mech. Res. Commun. 89
(2018) 1e5.
[39] M.S. Ansari, M. Aghaie-Khafri, Predicting flow localization in semi-solid
deformation, Int. J. Material Form. 11 (2) (2018) 165e173.
[40] O. Ludwig, J.M. Drezet, C.L. Martin, M. Suery, Rheological behavior of Al-Cu
alloys during solidification: constitutive modeling, experimental identifica-
tion, and numerical study, Metall. Mater. Trans. 36a (6) (2005) 1525e1535.
[41] O. Ludwig, J.M. Drezet, P. Meneses, C.L. Martin, M. Suery, Rheological
behavior of a commercial AA5182 aluminum alloy during solidification,
Mater. Sci. Eng., A 413 (2005) 174e179.
[42] T. Koshikawa, M. Bellet, C.A. Gandin, H. Yamamura, M. Bobadilla, Experi-
mental study and two-phase numerical modeling of macrosegregation
induced by solid deformation during punch pressing of solidifying steel in-
gots, Acta Mater. 124 (2017) 513e527.
[43] T. Kajitani, J.M. Drezet, M. Rappaz, Numerical simulation of deformation-
induced segregation in continuous casting of steel, Metall. Mater. Trans. 32
(6) (2001) 1479e1491.
[44] C.S. Li, B.G. Thomas, Thermomechanical finite-element model of shell
behavior in continuous casting of steel, Metall. Mater. Trans. B 35 (6) (2004)
1151e1172.
[45] J. Domitner, M.H. Wu, A. Kharicha, A. Ludwig, B. Kaufmann, J. Reiter,
T. Schaden, Modeling the effects of strand surface bulging and mechanical
soft reduction on the macrosegregation formation in steel continuous cast-
ing, Metall. Mater. Trans. 45a (3) (2014) 1415e1434.
[46] A.B. Phillion, S.L. Cockcroft, P.D. Lee, A three-phase simulation of the effect of
microstructural features on semi-solid tensile deformation, Acta Mater. 56
(16) (2008) 4328e4338.
[47] M. M'Hamdi, A. Mo, H.G. Fjaer, TearSim: a two-phase model addressing hot
tearing formation during aluminum direct chill casting, Metall. Mater. Trans.
37a (10) (2006) 3069e3083.
[48] S. Vernede, J.A. Dantzig, M. Rappaz, A mesoscale granular model for the
mechanical behavior of alloys during solidification, Acta Mater. 57 (5) (2009)
1554e1569.
[49] C. O'Sullivan, Particulate Discrete Element Modelling: a Geomechanics
Perspective, Spon, London, 2011.
[50] H. Shimizu, S. Murata, T. Ishida, The distinct element analysis for hydraulic
fracturing in hard rock considering fluid viscosity and particle size distri-
bution, Int. J Rock. Mech. Min. 48 (5) (2011) 712e727.
[51] D.F. Boutt, B.K. Cook, B.J.O.L. McPherson, J.R. Williams, Direct simulation of
fluid-solid mechanics in porous media using the discrete element and
lattice-Boltzmann methods, J. Geophys. Res.-Sol. Ea. 112 (B10) (2007).
[52] G.W. Bergantz, J.M. Schleicher, A. Burgisser, Open-system dynamics and
mixing in magma mushes, Nat. Geosci. 8 (10) (2015) 793e796.
[53] A. Leonardi, F.K. Wittel, M. Mendoza, H.J. Herrmann, Coupled DEM-LBM
method for the free-surface simulation of heterogeneous suspensions,
Comput. Part. Mech. 1 (1) (2014) 3e13.
[54] V. Slowik, J.W. Ju, Discrete modeling of plastic cement paste subjected to
drying, Cement Concr. Compos. 33 (9) (2011) 925e935.
[55] C.L. Martin, D. Bouvard, S. Shima, Study of particle rearrangement during
powder compaction by the discrete element method, J. Mech. Phys. Solid. 51
(4) (2003) 667e693.
[56] H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Discrete particle simulation of par-
ticulate systems: a review of major applications and findings, Chem. Eng. Sci.
63 (23) (2008) 5728e5770.
[57] B.K. Cook, D.R. Noble, J.R. Williams, A direct simulation method for particle-
fluid systems, Eng. Comput. 21 (2e4) (2004) 151e168.
[58] D. Noble, J. Torczynski, A lattice-Boltzmann method for partially saturated
computational cells, Int. J. Mod. Phys. C 9 (08) (1998) 1189e1201.
[59] Y.H. Han, P.A. Cundall, Resolution sensitivity of momentum-exchange and
immersed boundary methods for solid-fluid interaction in the lattice Boltz-
mann method, Int. J. Numer. Methods Fluid. 67 (3) (2011) 314e327.
[60] Y.H. Han, P.A. Cundall, LBM-DEM modeling of fluid-solid interaction in
porous media, Int. J. Numer. Anal. Model. 37 (10) (2013) 1391e1407.
[61] D.H. Johnson, F. Vahedifard, B. Jelinek, J.F. Peters, Micromechanics of un-
drained response of dilative granular media using a coupled DEM-LBM
model: a case of biaxial test, Comput. Geotech. 89 (2017) 103e112.
[62] M. Sistaninia, A.B. Phillion, J.M. Drezet, M. Rappaz, Simulation of semi-solid
material mechanical behavior using a combined discrete/finite element
method, Metall. Mater. Trans. 42A (1) (2011) 239e248.
[63] M. Sistaninia, S. Terzi, A.B. Phillion, J.M. Drezet, M. Rappaz, 3-D granular
modeling and in situ X-ray tomographic imaging: a comparative study of hot
tearing formation and semi-solid deformation in Al-Cu alloys, Acta Mater. 61
(10) (2013) 3831e3841.
[64] L. Yuan, C. O'Sullivan, C.M. Gourlay, Exploring dendrite coherency with the
discrete element method, Acta Mater. 60 (3) (2012) 1334e1345.
[65] K. Umetani, K. Uesugi, M. Kobatake, A. Yamamoto, T. Yamashita, S. Imai,
Synchrotron radiation microimaging in rabbit models of cancer for preclin-
ical testing, Nucl. Instrum. Methods Phys. Res. 609 (1) (2009) 38e49.
[66] D. Swinehart, The Beer-Lambert law, J. Chem. Educ. 39 (7) (1962) 333.
[67] S.R. Sternberg, Biomedical image-processing, Computer 16 (1) (1983) 22e34.
[68] H. Yasuda, T. Nagira, M. Yoshiya, N. Nakatsuka, A. Sugiyama, K. Uesugi,
K. Umetani, Development of X-ray imaging for observing solidification of
carbon steels, ISIJ Int. 51 (3) (2011) 402e408.
[69] S. Ganesan, D.R. Poirier, Densities of aluminum-rich aluminum-copper alloys
during solidification, Metall. Trans. A 18 (4) (1987) 721e723.
[70] J.L. Murray, The aluminium-copper system, Int. Met. Rev. 30 (1) (1985)
 T.C. Su et al. / Acta Materialia 163 (2019) 208e225
211e234.
[71] J.H. Hubbell, S.M. Seltzer, Tables of X-Ray Mass Attenuation Coefficients and
Mass Energy-absorption Coefficients from 1 KeV to 20 MeV for Elements Z ¼
1 to 92 and 48 Additional Substances of Dosimetric Interest. Radiation
Physics Division, PML, NIST, 1996.
[72] J.A. Dantzig, M. Rappaz, Solidification, EPFL Press, London, 2009.
[73] S. Ganesan, D.R. Poirier, Conservation of mass and momentum for the flow of
interdendritic liquid during solidification, Metall. Trans. B 21 (1) (1990)
173e181.
[74] L.E. Malvern, Introduction to the Mechanics of a Continuous Medium,
Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1969.
[75] M. Bornert, F. Bremand, P. Doumalin, J.C. Dupre, M. Fazzini, M. Grediac,
F. Hild, S. Mistou, J. Molimard, J.J. Orteu, L. Robert, Y. Surrel, P. Vacher,
B. Wattrisse, Assessment of digital image correlation measurement errors:
methodology and results, Exp. Mech. 49 (3) (2009) 353e370.
[76] F.J. Yang, X.Y. He, C.G. Quan, Characterization of dynamic microgyroscopes by
use of temporal digital image correlation, Appl. Opt. 45 (30) (2006)
7785e7790.
[77] M. Oda, H. Kazama, Microstructure of shear bands and its relation to the
mechanisms of dilatancy and failure of dense granular soils, Geotechnique
48 (4) (1998) 465e481.
[78] Y.H. Wang, S.C. Leung, A particulate-scale investigation of cemented sand
behavior, Can. Geotech. J. 45 (1) (2008) 29e44.
[79] S. Nemat-Nasser, Averaging theorems in finite deformation plasticity, Mech.
Mater. 31 (8) (1999) 493e523.
[80] E. Masad, B. Muhunthan, Three-dimensional characterization and simulation
of anisotropic soil fabric, J. Geotech. Geoenviron. 126 (3) (2000) 199e207.
[81] A. Drescher, G.D.J. De Jong, Photoelastic verification of a mechanical model
for the flow of a granular material, J. Mech. Phys. Solid. 20 (5) (1972)
337e340.
[82] K. Iwashita, M. Oda, Rolling resistance at contacts in simulation of shear band
development by DEM, J. Eng. Mech.-Asce 124 (3) (1998) 285e292.
[83] J.K. Morgan, M.S. Boettcher, Numerical simulations of granular shear zones
using the distinct element method - 1. Shear zone kinematics and the
micromechanics of localization, J. Geophys Res-Sol Ea 104 (B2) (1999)
2703e2719.
[84] M.J. Jiang, H.B. Yan, H.H. Zhu, S. Utili, Modeling shear behavior and strain
localization in cemented sands by two-dimensional distinct element method
analyses, Comput. Geotech. 38 (1) (2011) 14e29.
[85] A. Cottrell, R. Stokes, Effects of temperature on the plastic properties of
aluminium crystals, in: Proceedings of the Royal Society of London A:
Mathematical, Physical and Engineering Sciences, The Royal Society, 1955,
pp. 17e34.
[86] J.J. Hu, J.Y. Zhang, Z.H. Jiang, X.D. Ding, Y.S. Zhang, S. Han, J. Sun, J.S. Lian,
225
Plastic deformation behavior during unloading in compressive cyclic test of
nanocrystalline copper, Mater. Sci. Eng., A 651 (2016) 999e1009.
[87] S. Ganesan, R. Speiser, D.R. Poirier, Viscosities of aluminum-rich AlCu liquid
alloys, Metall. Trans. B 18 (2) (1987) 421e424.
[88] Y. Han, P. Cundall, Verification of two-dimensional LBM-DEM coupling
approach and its application in modeling episodic sand production in
borehole, Petroleum 3 (2) (2017) 179e189.
[89] K.K. Soundararajan, Multi-scale Multiphase Modelling of Granular Flows,
PhD Thesis, University of Cambridge, 2015.
[90] K. Kumar, J.Y. Delenne, K. Soga, Mechanics of granular column collapse in
fluid at varying slope angles, J. Hydrodyn. 29 (4) (2017) 529e541.
[91] X.Y. He, Q.S. Zou, L.S. Luo, M. Dembo, Analytic solutions of simple flows and
analysis of nonslip boundary conditions for the lattice Boltzmann BGK
model, J. Stat. Phys. 87 (1e2) (1997) 115e136.
[92] P. Lallemand, L.S. Luo, Lattice Boltzmann method for moving boundaries,
J. Comput. Phys. 184 (2) (2003) 406e421.
[93] PFC-particle Flow Code 5.0, Itasca Consulting Group, Inc., Minneapolis, USA,
2014.
[94] D. Markauskas, R. Kacianauskas, Investigation of rice grain flow by multi-
sphere particle model with rolling resistance, Granul. Matter 13 (2) (2011)
143e148.
[95] B. Meylan, S. Terzi, C.M. Gourlay, M. Suery, A.K. Dahle, Development of shear
bands during deformation of partially solid alloys, Scripta Mater. 63 (12)
(2010) 1185e1188.
[96] M.J. Jiang, H.S. Yu, D. Harris, Discrete element modelling of deep penetration
in granular soils, Int. J. Numer. Anal. Model. 30 (4) (2006) 335e361.
[97] S.A. Metz, M.C. Flemings, Hot tearing in cast metals, AFS Trans 77 (1969)
329e334.
[98] X.S. Li, Y.F. Dafalias, Dilatancy for cohesionless soils, Geotechnique 50 (4)
(2000) 449e460.
[99] L. Rothenburg, N.P. Kruyt, Critical state and evolution of coordination num-
ber in simulated granular materials, Int. J. Solid Struct. 41 (21) (2004)
5763e5774.
[100] S. Lobo-Guerrero, L.E. Vallejo, Discrete element method evaluation of gran-
ular crushing under direct shear test conditions, J. Geotech. Geoenviron. 131
(10) (2005) 1295e1300.
[101] S. Lobo-Guerrero, L.E. Vallejo, DEM analysis of crushing around driven piles
in granular materials, Geotechnique 55 (8) (2005) 617e623.
[102] L. Arnberg, G. Chai, L. Backerud, Determination of dendritic coherency in
solidifying melts by theological measurements, Mater. Sci. Eng., A 173 (1e2)
(1993) 101e103.
[103] O. Nielsen, L. Arnberg, A. Mo, H. Thevik, Experimental determination of
mushy zone permeability in aluminum-copper alloys with equiaxed mi-
crostructures, Metall. Mater. Trans. 30 (9) (1999) 2455e2462.